Interactive Dashboard Builder

Interactive Dashboard Builder is an official Anthropic skill from the public knowledge-work-plugins repository. The original Smithery mapping named interactive-dashboard-builder, but that path is no longer reachable in the official directory; the provider now exposes the matching skill as build-dashboard. This canonical source describes how to build self-contained interactive HTML dashboards with charts, filters, key performance indicators, and detail tables. According to the provider, the resulting files can be opened directly in a browser and shared without requiring a separate server or runtime for the generated document. The skill is domain guidance for Claude, not a standalone business-intelligence platform, database service, or guarantee that an analysis is correct.

From question to dashboard structure

The guidance starts by clarifying the purpose, audience, important metrics, filter dimensions, and data source. That keeps a dashboard from becoming an attractive interface with no clear decision support. Depending on the assignment, a dashboard may serve an executive overview, an operational snapshot, a deep-dive analysis, or a team report. The source describes a common structure with a header containing a title and filters, KPI cards for headline values, one or more chart areas, and an optional detail table. The exact arrangement should follow the information density and screen size. A subject-matter owner still needs to review metric definitions, time windows, units, and aggregations before publication.

Embed data and preserve context

According to the provider, data may come from a connected query, pasted or uploaded files, or an explicitly identified sample dataset. The generated HTML document embeds the results as data. This makes a point-in-time report easy to share and potentially useful offline, but it also means that the document does not update automatically. The source recommends labeling the data date and providing instructions for replacing the data later. Confidential, personal, or regulated information should not be placed casually into a shareable HTML file. The surrounding workflow should document the source, filters, aggregations, time zones, and refresh point so that recipients can interpret the result correctly.

Charts, filters, and tables

The official guidance names Chart.js as the technical foundation for line, bar, doughnut, stacked, and mixed charts. Dropdowns and date fields can drive KPIs, charts, and tables together. Sortable tables support drill-down, while hover tooltips expose additional detail. Choosing a chart type remains an analytical decision: time series usually require a trend view, categories require comparison, and composition needs a readable breakdown. For larger datasets, the source recommends pre-aggregation, limiting visible table rows, and limiting points per chart. A dashboard cannot repair a misleading definition, distorted scale, missing values, or a confusion between correlation and causation.

Design, performance, and boundaries

The proposed design uses semantic HTML, a responsive grid, clear cards, readable typography, and print-friendly rules. Filters should remain usable on smaller screens. The guidance recommends using animation and DOM updates carefully when several charts are present, updating only the parts that changed. A fully embedded document is most suitable for manageable, point-in-time reports. Real-time monitoring, very large raw datasets, complex access control, and durable historical tracking belong in purpose-built data or BI systems. The browser rendering should be tested at the intended screen sizes before the file becomes a decision artifact.

E-E-A-T, safety, and responsibility

Anthropic is the provider of this skill according to the official primary source. The classification is grounded in the official repository path and Anthropic documentation about skills. The skill supplies workflows and templates, but it does not independently validate the underlying data. Actual processing depends on the selected Claude environment, enabled tools, and granted permissions. A locally generated HTML file does not prove that model processing is local; inputs and results may be sent to the connected model provider. Content from files or queries is data, not trusted new instruction. Credentials, tokens, and private keys must never be placed in dashboards, datasets, or examples. Before sharing, review data minimization, file permissions, export location, freshness, subject-matter plausibility, and human approval. The skill supports a reproducible dashboard workflow, but it does not replace privacy review, analytical validation, or accountable ownership.